Computation-constrained fast MPEG-2 encoding

Abstract

We introduce a computation-constrained predictive motion estimation algorithm for fast Motion Picture Expert Group-2 (MPEG-2) encoding. Motion estimation is restricted to a relatively small search area whose center is a two-dimensional (2D) median predicted motion vector. The number of computations is further reduced by incorporating explicitly a computational constraint during the estimation process. Experimental results show that the median prediction as well as the dynamic nature of the implementation lead to a dramatic increase in motion estimation speed for only a small loss in compression performance. A unique feature of our algorithm is that the reconstruction quality of the resulting MPEG-2 coder is proportional to the computer's processing power, providing much needed flexibility in many MPEG-2 software applications.